Project 22: “Development and Tes6ng of New Tools”
Developing and Tes/ng Improved Tools for Power
System Planning and Opera/on under Uncertainty
Ray Zimmerman, Carlos Murillo-‐Sánchez,
Alberto Lamadrid, Daniel Muñoz-‐Álvarez, Tim Mount, Bob Thomas
CERTS Review, Cornell University
August 4-‐5, 2014
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
2
Tools Overview
3
MATPOWER
E4ST* based on DC version
Single Period SuperOPF (1st gen)
AC and DC versions
MOPS** DC version
MulN-‐period SuperOPF (2nd gen)
DC version + AC prototype
extensible OPF architecture high performance solvers
single-period explicit contingencies
stochastic cost endogenous reserves
environmental costs optimal investment
unit commitment
multi-period storage
flexible demand ramping
* E4ST – Engineering, Economic, Environmental Electricity Simulation Tool, formerly SuperOPF Planning Tool. ** MOPS – MATPOWER Optimal Power Scheduler, based on Multi-period SuperOPF with Unit Commitment (3rd generation)
new expansion planning tool AC and DC versions
parallel decomposition integer investment decisions
zonal elastic fuel supply finer time granularity
MOPS-‐AC AC prototype
parallel decomposition
2nd order update info
tesNng and simulaNon frameworks
MOPS** DC version
Background -‐ MoNvaNon • Tool Development
– Improve on the so\ware tools in current use for planning and operaNon of electric power systems, especially in light of industry trends:
• uncertainty (renewables, environmental regulaNon, etc.) • new technologies (storage, demand side parNcipaNon, microgrids, etc.)
• Tool TesNng – Demonstrate and measure the benefits of our approach over current system
operator pracNce. • Compare stochasNc approach to tradiNonal determinisNc approach using MOPS. • Compare receding horizon structure to day-‐at-‐a-‐Nme planning using MOPS.
• Impacts – Small improvements in efficiency can have significant economic and reliability
impacts. – Open source tools have far-‐reaching impact on research beyond this program.
4
Background -‐ Context • Tool Development
– MOPS • design does not preclude AC network model • co-‐opNmizing energy, endogenous reserves for conNngencies and load-‐following • internalized ramping costs • scenario tree recombinaNon, preserves mulNstage structure of decisions
– new expansion planning tool (basis for E4ST v2) • AC or DC network model • binary investment/reNrement decisions and costs • elasNcity of fuel supply • more Nme granularity
• Tool TesNng – performance tested in context of original problem with non-‐anNcipaNvity, not
approximaNons made to make problem tractable (mulN-‐stage decisions vs. two-‐stage)
5
Background -‐ Status • Tool Development
– MATPOWER • released versions 5.0 and 5.1 in past year • expect version 6.0 with MOPS in a few months • moving toward public open source project with open development paradigm
– E4ST • new web site (Schulze project) to host data, tools, results • version 1 of core solver so\ware stable
– public distribuNon requires addiNonal cleanup and documentaNon
– MOPS • being acNvely used for tesNng • integraNon into MATPOWER 6, requires more documentaNon, cleanup of code • receding horizon requires further modificaNons (parNally completed)
– next gen planning tool, foundaNon for E4ST v2 • prototype complete, being used in tesNng • integrate into MATPOWER or separate E4ST v2 project when ready
6
Background -‐ Status • Tool Development (conNnued)
– AC version of MOPS • current prototype not ready for prime Nme due to convergence challenges • explore Newton-‐based coordinaNon updates to address convergence problems
• Tool TesNng – integrated simulaNon plaeorm for MOPS tesNng, two-‐seflement, receding
horizon, etc. • original afempt at grand unified simulator bogged down in details • two seflement simulator running • receding horizon delayed
– two stage framework, stochasNc vs. determinisNc • lots of obstacles in design of 118 bus comparisons, calibraNon of inputs, structure of simulaNon • comparisons almost complete, expect paper submission in Aug or Sep
– receding horizon • required addiNonal changes to MOPS (nearly complete) • requires more general simulator, pursuing gehng a student to help with this • addiNonal challenges expected as we move to generaNng wind scenarios on the fly
7
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
8
MATPOWER Free, open-‐source power system simulaNon environment with extensible OPF and interfaces to state-‐of-‐the-‐art solvers.
9
hfp://www.pserc.cornell.edu/matpower/ – used worldwide in teaching, research,
industry – momentum & impact conNnues to grow
• 1062 cita/ons of 2 main MATPOWER papers*
• 664 cita/ons of MATPOWER so\ware/manual*
– serves as foundaNon for all tools in this project
R. D. Zimmerman, C. E. Murillo-‐Sánchez, and R. J. Thomas, “MATPOWER Steady-‐State OperaNons, Planning and Analysis Tools for Power Systems Research and EducaNon,” Power Systems, IEEE Transac6ons on, vol. 26, no. 1, pp. 12-‐19, Feb. 2011. * Google Scholar, 8/3/15
Annual MATPOWER Downloads
10
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Annual MATPOWER Downloads by Version
5.1
5.0
4.1
4.0
3.2
3.0
2.0
1.0
0
2000
4000
6000
8000
10000
12000
14000
16000
18000
20000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Annual MATPOWER Downloads by Version
5.1
5.0
4.1
4.0
3.2
3.0
2.0
1.0
CumulaNve MATPOWER Downloads
11
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Cumula/ve MATPOWER Downloads by Version
5.1
5.0
4.1
4.0
3.2
3.0
2.0
1.0
0
10000
20000
30000
40000
50000
60000
70000
80000
90000
100000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Cumula/ve MATPOWER Downloads by Version
5.1
5.0
4.1
4.0
3.2
3.0
2.0
1.0
MATPOWER Releases – v5.0 • Version 5.0b1 – released July 1, 2014
– conNnuaNon power flow – applicaNon of SDP relaxaNons of PF equaNons – extensible opNons architecture – tools for detailed reporNng of case data, connecNvity, manipulaNng islands – PSS/E RAW import capability – new and updated support for 3rd party solvers
• Version 5.0 (final) – released Dec 17, 2014 – enhanced PSS/E RAW import, more robust, support for more versions – so\ limits on DC OPF branch flows – more user sefable parameters for default interior point solver – performance enhancements – bug fixes
12
• Version 5.1 (final) – released Mar 20, 2015 – new license
• switched to more permissive 3-‐clause BSD license from GPL v3
– new case files • four new case files represenNng parts of European high voltage grid • models ranging from 89 to 9421 buses (largest system distributed with MATPOWER)
– new documentaNon • on-‐line funcNon reference at hfp://www.pserc.cornell.edu/matpower/docs/ref/
– new features • unified interface for 3rd party mixed integer solvers for MILP/MIQP • support for PARDISO as linear solver used by interior point algorithm for AC OPF • new and updated support for 3rd party solvers, including OPTI Toolbox (CLP, GLPK, IPOPT),
IPOPT-‐PARDISO • network reducNon toolbox (Tylavsky, Zhu)
– performance enhancements – other refinements and bug fixes
13
MATPOWER Releases – v5.1
New License BSD or MIT style license • permissive • short and to the point • lets people do what they want, as
long as they – provide afribuNon – don’t hold you liable
• leaves open commercializaNon opNons
• E.g. FreeBSD, jQuery, Rails
GPL license • copyle\ • legally complex • requires distribuNon of any
modificaNons or derivaNves to be under same terms
• designed to keep research results from transiNoning to proprietary products
• E.g. Linux, Git, WordPress
14 Sources: • Why you should use a BSD style license for your Open Source Project, http://www.freebsd.org/doc/en_US.ISO8859-1/articles/bsdl-gpl/article.html • http://choosealicense.com/
PARDISO – Large-‐scale AC OPF • Bofleneck in primal-‐dual interior point solvers (IPOPT, Knitro, MIPS) is the Newton
update step, i.e. solving Ax = b • PARDISO – hfp://www.pardiso-‐project.org
– thread-‐safe, high-‐performance, robust, memory efficient so\ware for solving large sparse linear systems of equaNons on shared/distributed-‐memory mulNprocessors
• Largest AC OPF model solved is 3000-‐bus Polish model with 63-‐conNngencies – roughly equivalent to 193,000-‐bus AC OPF, plus extras – size previously limited by RAM (64 GB on 12-‐core machine)
• When using MIPS, switching from Matlab’s built-‐in x=A\b to PARDISO results in … – 30x speedup on 12-‐core machine – order of magnitude decrease in RAM requirement
• IPOPT-‐PARDISO – currently MATPOWER’s fastest AC OPF solver – solves this case in ~18 minutes on my laptop (2014 MacBook Pro) – developers looking for large-‐scale systems to test their solver, MATPOWER integraNon – IPOPT-‐PARDISO distributed on PARDISO site under “MATPOWER libraries” – MATPOWER is helping to drive advances in high-‐performance linear system solvers
15
STAC
16
STAC
17
MATPOWER Development – v6.0-‐dev • ZIP load model
– experimental feature based on contributed code • performance enhancements
– large speedups when running many small problems
• MOPS integraNon (Ray) – code cleanup
• GNU Octave compaNbility • splihng program opNons from input data structures • adding UC to supporNng code for data input, auto-‐generaNon of sensible default data
– documentaNon, manuals and in code – automated tests – tutorial examples
• MOPS development – beyond MATPOWER 6 (Carlos) – probabilisNc iniNal state features needed for receding horizon applicaNon – AC prototype convergence improvement
18
MATPOWER Project DirecNons • First steps to address need for a sustainable long-‐term plan. • Apply for 3 years of funding through NSF SI2 (So\ware Infrastructure for
Sustained InnovaNon) program. – Goals include to “support the crea/on and maintenance of an innovaNve, integrated, reliable,
sustainable and accessible soQware ecosystem providing new capabiliNes that advance and accelerate scien/fic inquiry and applicaNon at unprecedented complexity and scale.”
– Division of Electrical, CommunicaNons and Cyber Systems “is parNcularly interested in proposals which provide wider, more flexible access to more advanced general algorithms in the areas of electronic and photonics device simulaNon (accounNng for quantum many body effects), computaNonal intelligence, nonlinear op/miza/on or energy system design.”
• Move MATPOWER to true public open source project/open dev paradigm – public code repository, mulNple commifers – public bug tracking facility, user/developer forums, improved web-‐site – core project documents defining project, goals, policies, how to contribute – streamlined “on ramps” for users and developers – effecNve process in place for incorporaNng contribuNons and feedback,
including from other CERTS projects 19
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
20
economicdispatch
add DCnetworkmodel
economicdispatch
DC OPF
add DCnetworkmodel
economicdispatch
add ACnetworkmodel
DC OPF
economicdispatch
AC OPF
add ACnetworkmodel
DC OPF
economicdispatch
add zonal reserves
AC OPF
DC OPF
economicdispatch
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves
add zonal reserves
AC OPF
DC OPF
economicdispatch
add contingencies
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
add contingencies
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
addstochastic
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
addstochastic
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch
MATPOWER
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
secureAC OPF
AC OPF
DC OPF
economicdispatch
1st gen SuperOPF
secureED
secureDC OPF
secureAC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticED
stochasticDC OPF
EDw/reserves
DC OPFw/reserves DC OPF
economicdispatch
MOPS
single deterministicsystem state
multiple probabilisticsystem states
secureED
secureDC OPF
secureAC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticsecureAC OPF
stochasticED
stochasticDC OPF
stochasticAC OPF
EDw/reserves
DC OPFw/reserves
AC OPFw/reserves AC OPF
DC OPF
economicdispatch secure
ED
secureDC OPF
stochastic secure
ED
stochasticsecureDC OPF
stochasticED
stochasticDC OPF
EDw/reserves
DC OPFw/reserves DC OPF
economicdispatch
MOPS
MOPS ConNnuous Single Period Problems
21
Tutorial Example System • 3-‐bus triangle network • generators
– 2 idenNcal 200 MW gens at bus 1, diff reserve cost – 500 MW gen at bus 2 – all 3 have idenNcal quadraNc generaNon costs
• load 450 MW at bus 3, curtailable @ $1000 • branches
– 300 MW limit, line 1–2 – 240 MW limit, line 1–3 – 300 MW limit, line 2–3
• adequacy requirement – reserve requirement
• 150 MW limit
– conNngencies • generator 2 at bus 1 • line 1–3
• wind – 100 MW unit at bus 2 – 3 samples of normal distribuNon around 50 MW
240 MW
bus 1 bus 2
bus 3
200 MW 200 MW 500 MW
450 MW
300 MW
300 MW
100 MW
gen 1 gen 2 gen 3
22
Single Period ConNnuous Examples
23
Economic Dispatch
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
24
Economic Dispatch w/reserves
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
25
DC OPF
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
26
DC OPF w/reserves
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
27
secure DC OPF
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
28
stochastic DC OPF
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
Single Period ConNnuous Examples
29
secure stochastic DC OPF
Gen1 2 3 4
MW
0
50
100
150
200
250
300Generation & Reserves
Bus1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Nodal Prices
Line1 2 3
$/MW
0
10
20
30
40
50
60
70
80
90
100Flow Constraint Shadow Prices
continuousdispatch
singleperiod
addintegercommitvars
continuousdispatch
singleperiod
with integercommitment
continuousdispatch
singleperiod
addstartup/shutdowncosts
with integercommitment
continuousdispatch
singleperiod
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
add multiple periods
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
add min
up/downtimes
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
add ramping
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
multi-period+ ramping
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
full UC + dispatchproblem
full look-aheaddispatch problem
multi-period+ ramping
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
multi-period+ ramping
addstorage
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
multi-period+ ramping
multi-period+ storage
multi-period+ ramping+ storage
with integer commitment+ transition costs
+ min up/down times
multi-period
with integercommitment + transition
costs
with integercommitment
continuousdispatch
singleperiod
MOPS Mixed Integer and MulN-‐Period Problems
30
Unit Commitment Examples
31
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
Base : No Network
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
Unit Commitment Examples
32
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
+ DC Network
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
Unit Commitment Examples
33
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
+ Startup/Shutdown Costs
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
Unit Commitment Examples
34
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
+ Min Up/Down Time Constraints
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
Unit Commitment Examples
35
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
+ Ramping Constraints/Ramp Reserve Costs
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
Unit Commitment Examples
36
Period1 2 3 4 5 6 7 8 9 10 11 12
Uni
t Com
mitm
ent
Gen 3 @ Bus 2
Gen 2 @ Bus 1
Gen 1 @ Bus 1
+ Storage
Period1 2 3 4 5 6 7 8 9 10 11 12
Gen
erat
ion,
MW
0
100
200
300
400
Period1 2 3 4 5 6 7 8 9 10 11 12N
odal
Pric
e, $
/MW
h
0
50
100
Net
Loa
d, M
W
0
100
200
300
400
MOPS Status
• Well on way to release in MATPOWER 6.0 – Need several more months to finalize code cleanup and documentaNon
• Beyond first public release – modificaNons for probabilisNc starNng point, required by receding horizon
– implement fix for stochasNc cost distorNon related to interacNon between commitment status and conNngency outages
– conNnued work on convergence of AC prototype
37
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
38
TesNng and SimulaNon Framework
39
modeling devices, networks,
storage, uncertainty, markets
problem formulaNon
data
algorithms
solvers
Simula/on Environment defines 6me structure, data/informa6on flow pa\erns between …
▸ mul6ple decision stages (e.g. day-‐ahead UC, 5-‐min dispatch/pricing) ▸ sequen6al solves of a given stage ▸ actual opera6on
TesNng MOPS • MOPS is a general purpose opNmizaNon tool
– like an OPF – can be used in many contexts
• To test MOPS, we need to define a context – create a plan given a model of forecasted uncertainty – update/execute that plan as uncertainty is revealed through Nme – measure performance (e.g. cost, other metrics) – Monte Carlo comparisons
• QuesNons – How to execute the plan?
• what is locked in, what’s free to change? – OperaNng policy vs. market design?
40
TesNng MOPS
Stochas/c Secure UC+OPF
• mulNple scenarios for
demand and renewable availability
• explicit conNngencies for security
Determinis/c UC+OPF
• single scenario with
expected demand and renewable availability
• zonal reserve requirements for security
41
Two Seflement Framework
• 1st seflement – solves a mulN-‐period plan resulNng in day-‐ahead commitment decisions and reserve allocaNons
• 2nd seflement – solves single-‐period problem to determine energy dispatch and conNngency reserve allocaNon subject to
• UC decisions from 1st seflement • dispatch from previous period 2nd seflement • newly revealed uncertainty
– currently using 2nd seflement to approximate actual operaNon
42
TesNng Structure
• Given: – historical temp, wind, demand up to operaNng day (any selected day of interest)
– ARIMA model of temp, wind, demand that can generate potenNal realizaNons of the operaNng day
• For each approach: 1. Solve 1st seflement problem for the day (based on
uncertainty predicted by the ARIMA model). 2. Select N realizaNons of the day generated by ARIMA
model, for each solve 2nd seflement problems sequenNally for each hour, subject to 1st seflement.
43
118-‐bus Test System
44
DC Network Example number of …
buses 118
convenNonal generators 42
wind farms 12
grid-‐level storage units 0
curtailable loads 99
periods in horizon, |T| 24
scenarios per period, |Jt| 5
conNngencies per scenario, |Ktj| – 1 7
variables in resulNng MIQP 582,990
constraints in resulNng MIQP 1,536,006
45
Typical Wind Trajectories Site 3
Hours0 5 10 15 20 25
Perc
enta
ge o
f Max
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Profiles vs. Trajectories, Site 3
46
47 Period
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Gen 42 @ Bus 116, coalGen 41 @ Bus 113, coalGen 40 @ Bus 111, coalGen 35 @ Bus 100, coalGen 33 @ Bus 91, coalGen 32 @ Bus 90, coalGen 31 @ Bus 89, ngGen 30 @ Bus 89, ngGen 29 @ Bus 87, coalGen 28 @ Bus 80, coalGen 27 @ Bus 77, ngGen 26 @ Bus 76, ngGen 25 @ Bus 73, coalGen 24 @ Bus 72, coalGen 23 @ Bus 69, ngGen 22 @ Bus 69, ngGen 21 @ Bus 66, ngGen 20 @ Bus 66, ngGen 19 @ Bus 65, ngGen 18 @ Bus 65, ngGen 17 @ Bus 62, ngccGen 16 @ Bus 61, ngccGen 15 @ Bus 59, ngccGen 14 @ Bus 54, ngGen 13 @ Bus 49, coalGen 12 @ Bus 49, coalGen 11 @ Bus 46, coalGen 10 @ Bus 42, coalGen 9 @ Bus 40, coalGen 8 @ Bus 27, coalGen 7 @ Bus 26, coalGen 6 @ Bus 25, coalGen 5 @ Bus 24, coalGen 4 @ Bus 12, coalGen 2 @ Bus 8, coalGen 1 @ Bus 6, ng
Unit Commitment - Stochastic
48 Period
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Gen 42 @ Bus 116, coalGen 41 @ Bus 113, coalGen 40 @ Bus 111, coalGen 35 @ Bus 100, coalGen 33 @ Bus 91, coalGen 32 @ Bus 90, coalGen 31 @ Bus 89, ngGen 30 @ Bus 89, ngGen 29 @ Bus 87, coalGen 28 @ Bus 80, coalGen 27 @ Bus 77, ngGen 26 @ Bus 76, ngGen 25 @ Bus 73, coalGen 24 @ Bus 72, coalGen 23 @ Bus 69, ngGen 22 @ Bus 69, ngGen 21 @ Bus 66, ngGen 20 @ Bus 66, ngGen 19 @ Bus 65, ngGen 18 @ Bus 65, ngGen 17 @ Bus 62, ngccGen 16 @ Bus 61, ngccGen 15 @ Bus 59, ngccGen 14 @ Bus 54, ngGen 13 @ Bus 49, coalGen 12 @ Bus 49, coalGen 11 @ Bus 46, coalGen 10 @ Bus 42, coalGen 9 @ Bus 40, coalGen 8 @ Bus 27, coalGen 7 @ Bus 26, coalGen 6 @ Bus 25, coalGen 5 @ Bus 24, coalGen 4 @ Bus 12, coalGen 2 @ Bus 8, coalGen 1 @ Bus 6, ng
Unit Commitment - Deterministic
49 Period
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24Gen 42 @ Bus 116, coalGen 41 @ Bus 113, coalGen 40 @ Bus 111, coalGen 35 @ Bus 100, coalGen 33 @ Bus 91, coalGen 32 @ Bus 90, coalGen 31 @ Bus 89, ngGen 30 @ Bus 89, ngGen 29 @ Bus 87, coalGen 28 @ Bus 80, coalGen 27 @ Bus 77, ngGen 26 @ Bus 76, ngGen 25 @ Bus 73, coalGen 24 @ Bus 72, coalGen 23 @ Bus 69, ngGen 22 @ Bus 69, ngGen 21 @ Bus 66, ngGen 20 @ Bus 66, ngGen 19 @ Bus 65, ngGen 18 @ Bus 65, ngGen 17 @ Bus 62, ngccGen 16 @ Bus 61, ngccGen 15 @ Bus 59, ngccGen 14 @ Bus 54, ngGen 13 @ Bus 49, coalGen 12 @ Bus 49, coalGen 11 @ Bus 46, coalGen 10 @ Bus 42, coalGen 9 @ Bus 40, coalGen 8 @ Bus 27, coalGen 7 @ Bus 26, coalGen 6 @ Bus 25, coalGen 5 @ Bus 24, coalGen 4 @ Bus 12, coalGen 2 @ Bus 8, coalGen 1 @ Bus 6, ng
Unit Commitment - Both
StochasticDeterministicBoth
50 Period
0 5 10 15 20 25
MW
0
500
1000
1500
2000
2500
3000
3500
4000
4500
5000Committed Capacity vs. Expected Net Demand
StochasticDeterministicExpected Net Demand
51
Stochastic Deterministic
Expe
cted
Cos
ts ($
)
#106
0
0.5
1
1.5
2
2.5
3Expected System Costs
FuelNo LoadUnit CommitmentDA ReserveRT ReserveLoad Not Served
Expected Cost Comparison
52
Stochas/c Determinis/c Difference
fuel $1,386,000 $1,564,000 9%
no load $449,000 $440,000 0%
UC $5,000 $9,000 0%
DA Reserve $17,000 $51,000 2%
RT Reserve $8,000 $45,000 2%
LNS $66,000 $650,000 30%
Total $1,931,000 $2,760,000 43%
53 Standard Deviation ($) #105
0 1 2 3 4 5 6
Aver
age
Cos
t ($)
#106
0
0.5
1
1.5
2
2.5
3
3.5
4Performance of Dispatch Policies
stochasticdeterministic
Two-‐Seflement Challenges • Began with the idea that 1st seflement contracts for commitment, energy,
reserves and ramping would provide “look-‐ahead view” to constrain single-‐period 2nd seflement problem – too restricNve – resulted in shedding load when unused capacity was available
• unused capacity hadn’t been contracted – consequence of
• myopic (single-‐period) second seflement problem • simplified uncertainty model
• Moved to having second seflement take only UC from first seflement results – full range of generator available for dispatch and reserves – conNnue to guard against conNngencies – ignores first seflement ranges that guarantee ramping feasibility
54
QuesNon
• How should we use the informaNon from the results of coarser, longer horizon “look-‐ahead” plan to guide the updaNng of that plan by a subsequent finer grain, but shorter horizon problem with new informaNon? – context of tesNng environment – market design context
55
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
56
Comparison to Current E4ST New Formula/on E4ST
AC or DC network model DC network model only
binary variables for investment/reNrement decisions conNnuous variables for investment/reNrement decisions
single Nme-‐linked opNmizaNon for enNre horizon independent sequenNal opNmizaNons
temporally co-‐opNmized investment/reNrement decisions independent sequenNal investment/reNrement decisions
fine Nme-‐granularity on investment decisions, yearly steps coarse Nme-‐granularity on investment decisions, decade steps
technology-‐specific invest-‐to-‐deploy delays uniform invest-‐to-‐deploy delay (granularity of investment cycles)
explicit zonal operaNng reserves availability factors as proxy for operaNng reserves, etc.
linear elasNc zonal fuel supply funcNons, possibly with delay exogenous fuel prices
poten6al to include ramping and UC via typical trajectories* operaNons consists of single independent hours
iteraNve soluNon of model decomposiNon direct soluNon of single large model**
highly parallelizable limited opportuniNes for parallel computaNon
approximate soluNon with small non-‐zero duality gap exact soluNon**
explicit hydro constraints, etc. not yet implemented* total output constraints for hydro, emissions, RPS
* future enhancement ** for each independent investment cycle
57
Integer Deployment States
58
τ
τ+1
3
2
1
τ
3
2
1
τ+1
τ+2 τ+2
t-1 t 0
0
0
Investment not approved U =0
Under construction U =0
Online U =1
Retired U =0
Operation cost
τ = Construction delay
SoluNon
• Lagrangian relaxaNon-‐based coordinaNon and separaNon of inner minimizaNons
• Two sets of minimizaNons – independent (AC) OPF problems
• one per hour type per interval of planning horizon – independent dynamic programs
• one per generator or project
• Highly separable, highly parallelizable • PotenNal challenge: LR convergence
59
Preliminary Test
60
Test Setup • Colombian system + several addiNonal projects
– 86 buses – 187 exisNng generaNon units
• zonal reserve requirement, 5 zones – 15% of local demand in each reserve zone
• potenNal new projects – 2 large hydro – 18 combined cycle gas – 5 large coal
• 5 hour types • 24 years • inelasNc fuel supply • 3% yearly discount rate
61
Online Status of Plants
62 Year
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24new coalnew coalnew coalnew coalnew coalnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew ngnew hydronew hydroexisting syncgensgen 160 : hydrogen 155 : hydrogen 154 : nggen 153 : hydrogen 142 : nggen 126 : hydrogen 123 : nggen 119 : nggen 118 : coalgen 116 : nggen 115 : coalgen 108 : nggen 106 : ngccgen 105 : nggen 104 : nggen 103 : nggen 102 : nggen 101 : coalgen 100 : ngccgen 96 : ngccgen 95 : ngccgen 94 : coalgen 85 : coalgen 84 : nggen 83 : nggen 82 : nggen 74 : ngccgen 62 : ngccgen 58 : ngccgen 57 : ngccgen 56 : ngccgen 55 : ngccgen 51 : ngccgen 50 : ngccgen 49 : ngccgen 48 : ngccgen 47 : ngccgen 38 : coalgen 31 : coalgen 30 : coalgen 29 : coalgen 28 : nggen 27 : nggen 18 : ngccgen 17 : ngccgen 16 : nggen 15 : ngccgen 14 : coalgen 13 : ngccgen 12 : ngccgen 11 : ngccgen 10 : ngccgen 6 : ngccmany existing hydro units
Expansion Plan
Dual variables
63
Preliminary Results
• most dual variables have sefled – need to explicitly compute duality gap
• new hydro plants and some combined cycle gas units selected for construcNon
• all natural gas units reNred immediately – NG is expensive, but units are required to cover hydro shortages during El Niño events
– requires explicit hydro constraint and possibly explicit modeling of uncertainty to capture properly
64
Status
• implementaNon of basic formulaNon (presented last year) completed
• new elasNc linear fuel supply funcNon – feature implementaNon completed – generators belong to (possibly overlapping) “fuel zones” in which they compete for fuel in given Nme horizon/season
• producNon costs can now include fixed cost proporNonal to installed capacity of project
• test study for Colombian system is undergoing calibraNon of model for fuel supply funcNon
65
To Do • Add seasonal output restricNons
– hydro constraints • currently requires manually adjusNng hydro generaNon cost
– emissions caps – RPS standards
• Coordinate with current E4ST users to close any other gaps and plan transiNon
• Future – Change scenarios from being “typical hours” to being “typical
trajectories” to model ramping requirements – Explore ways to include transmission expansion and uncertainty.
66
Overview
• Background – moNvaNon, context, status • MATPOWER – updates and direcNon • MOPS – MATPOWER OpNmal Power Scheduler • Tes/ng MOPS – simulaNon framework, tests • New expansion planning tool • AC convergence of MOPS
67
DecomposiNon of AC MOPS
68
single large LP/QP
Convergence of AC MOPS con6nuous case
• current LR coordinaNon with subgradient-‐based lambda update – slow and unreliable, especially when some states have negaNve prices
• improved update strategy to accelerate convergence – augment Lagrangian with squares of AC flow equaNons
• introduces 2nd order derivaNves of injecNons – should improve homing capability of lambda updates
69
Second Order Update Strategy
• Perform lambda update from sensiNvity analysis of FOC of augmented Lagrangian – similar to method of mulNpliers – requires solving huge linear system
• order-‐of-‐magnitude speed-‐ups in solving Ax = b (PARDISO) • lambda update based on 2nd order info seems doable scale-‐wise
• Status – under implementaNon (currently coding Hessians)
70
Support for Other CERTS Projects
• Bill Schulze – conNnued E4ST applicaNon/enhancements
• Ben Hobbs – integraNon with E4ST • Lindsay Anderson – integraNon of chance constrained approach with SuperOPF/MOPS
• Zhifang Wang – integraNon of syntheNc power grid modeling capability into MATPOWER
• HyungSeon Oh – AC OPF enhancements • Kory Hedman – stochasNc unit commitment
71
Summary of FY15 • Complete current MOPS tes/ng on two-‐seYlement structure and
submit paper to IEEE TransacNons. • Complete MOPS integra/on and release MATPOWER 6. • Submit NSF grant to move MATPOWER project to public open
development paradigm as first step toward sustainable long-‐term plan for MATPOWER.
• Finish building simulator, compleNng receding horizon comparisons. • Integrate new generaNon expansion planning tool (E4ST v2) into
MATPOWER suite. • Explore a Newton-‐based coordinaNon scheme to resolving AC MOPS
convergence issues. • Support other projects using MATPOWER/SuperOPF tools and
frameworks.
72
QuesNons?
73